If there is traffic to fill up blocks to that size, and the blocks cannot grow to that size, then the system fails.
Your lack of faith, imagination, and economic literacy does not constitute an emergency for Bitcoin.
Bitcoin is antifragile. That means rather than fail in the face of adversity it grows stronger.
When blocks grow to >1MB, less important (IE valuable) transactions will simply move to other less cluttered blockchains.
This is Econ 101. This is Julian Simon winning The Bet against doom-mongering catastrophist Paul Ehrlich. This is the Substitution Effect, powered by the invisible hand.
Your Maximalist Monopolist conceit that Bitcoin is the only available blockchain in the entire universe (and thus must be all-inclusive) is demonstrably false.
Bitcoin The Canonical Reference Implementation's job is to replace the BIS, not fucking Visa, Paypal, Western Union, and little Sally's piggy bank.
The BIS effects every coffee purchase in the world, but not directly. Bitcoin, as the new digital gold standard, is destined to do the same job, only better.
Individual central/local banks and payment networks will be replaced by any combination of sufficiently secured blockchains (merged mined and/or using different POWs). The dark pools will be denominated in Monero. Bitcoin isn't a one-size-fits-all solution for everyone and everything, although its blockchain technology will be extended for those purposes.
A $2 coffee or $20 remittance to Africa doesn't belong in the same dataset as billion dollar trans-institutional settlements. If you insist they must cohabitate, Bitcoin will collapse under its own weigh, a hypertrophic dinosaur victim of its own success ready to die in the nearest tar pit or meteor crater.
At most, blocksize should double when block reward halves. Anything more is an inflationary free-rider-subsidizing Bloatcoin/Gavincoin/GigaCoin/VisaCoin alt-fork, not BTC proper.
To believe that trade-offs do not exist and 20MB blocks are somehow magically free is also economic illiteracy. Econ 101 tells us if you raise the cost of full nodes, you will have less of them.
Expensive/limited bandwidth, not cheap disk space, is the problem (regardless of what RoadStress the idiot says). The present ad-hoc attempts to subsidize nodes are a glaring admission of market failure, which will be greatly exacerbated by 20MB+ blocks.
Gavin's 20MB proposal is 100% textbook 'Embrace, Extend, Extinguish.' Specifically, the 'Extend' phase. And then comes the axe, as slow/hardened connections cannot propagate the bloat and nodes outside of Equation Group-vulnerable datacenters go dark, tearing holes in our mesh until only a fragile patchwork remains.
Keep Bitcoin Elite!TM